Agentic AI Governance: Data Foundations & Real-Time Policy Enforcement

Session Track: Graph + AI in Production

Session Time: 09:30 am - 10:00 am April 15

Session description

The surge toward agentic AI demands an architectural response to the inevitable challenges of control, governance, and compliance at scale. This session will detail the engineering discipline required to build trustworthy, production-ready AI. You will learn to frame governance not as compliance, but as a technical stack built on a robust Data Foundation and Real-Time Policy Enforcement.

Adobe Experience Platform (AEP) enforces data usage policies through a governance graph built on Neo4j, evaluating policy checks at very high volume across complex entity hierarchies, ensuring data usage remains compliant in real-time.

In this session, we will present two innovations based on this architecture.

1. Data Foundations: You will learn how Adobe migrated its governance architecture from a document-centric (Cosmos DB) to a property-centric graph model (Neo4j), transforming governance labels and policies into first-class graph nodes.

2. Real-Time Policy Enforcement: We will walk through an agentic AI system that translates natural language questions into production-grade Cypher queries, executes them against governance graphs, and explains the results. You will learn patterns for schema-aware prompt engineering for reliable LLM-generated Cypher at production scale and multi-tool agentic workflows (LLM + Neo4j + external context) for graph debugging.

You will learn:
• How to model governance policies as property-centric graphs (not entity-centric documents)
• Schema-aware prompt engineering for reliable LLM-generated Cypher at production scale
• Multi-tool agentic workflows (LLM + Neo4j + external context) for graph debugging
• Real-world patterns for deep inheritance traversals and policy evaluation at scale